It’s nice to see social media bringing out the best in people.
There has been a lot of talk in recent weeks about prediction markets–either prediction markets inflating the suggested odds of extremely improbable candidates–or whales purportedly manipulating certain markets, such as contracts tied to Trump winning. Or the apparent disconnect between prediction market prices vs. polls, in which Trump’s odds appear conspicuously high relative to how well he is polling. All of these can be explained non-nefarious factors that does not involve collusion/manipulation of markets.
First item: A Mystery $30 Million Wave of Pro-Trump Bets Has Moved a Popular Prediction Market:
Vice President Kamala Harris and former President Donald Trump are neck and neck in the polls. But in one popular betting market, the odds have skewed heavily in Trump’s favor, raising questions about a recent flurry of wagers and who is behind them.
Over the past two weeks, the chances of a Trump victory in the November election have surged on Polymarket, a crypto-based prediction market. Its bettors were giving Trump a 62% chance of winning on Thursday, while Harris’s chances were 38%. The candidates were in a dead heat at the start of October.
It’s possible for whales to collude to inflate the price of prediction markets to either engender some sort of self-fulfilling prophecy or feedback effect, or because they have insider information (e.g. October surprise). But it would not make financial sense for whales to collude just for the sake of moving the market. Sure, whales could temporarily distort the market by clearing out the order book, but then this would quickly be reverted as arbitrageurs on the sidelines issue new contracts and equilibrize the market, to profit from this temporary distortion. At the same time, earlier buyers would liquidate their inflated contracts for a quick profit, adding to the selling pressure. For a market as liquid and widely-followed as contracts tied the outcome of 2024 election, prices should not remain distorted for very long.
There is a less sinister explanation, which is that polls should not be treated as probabilities. In regard to Kamala, prediction markets will tend to discount her odds relative to polls. Why? Because the Electoral College and the ‘surprise’ factor works to Trump’s favor, like in 2016. Or Bush in 2000, who also won despite losing the popular vote. Polls may not account or capture this lumpiness of the Electoral math or uncertainty of turnout as well as prediction markets do. So if poll aggregators, such as 538, show a 50% favorability for Kamala, this converts to a less-than-50% chance of winning due to the aforementioned factors, giving a contract value of perhaps only 40 cents for ‘yes’ instead of 50 cents. 40% is the closer to the true probability or her winning, not 50%.
Second Item: inflated probabilities for improbable events:
Reminder that Polymarket gave Michelle Obama a 10% chance of being the Democrat nominee earlier this year. A lot of money is no guarantee of rationality. This was the clearest example of this point I’ve ever seen.
— Richard Hanania (@RichardHanania) October 22, 2024
It may seem irrational for the RFK ‘yes’ contract to have traded traded at 3 cents, suggesting a 3% probably of RFK winning the 2024 election, when his odds were otherwise vanishingly small–way lower than 3%. Same for the ‘no’ contract trading at 97 cents instead of $1.00. Or Michelle Obama having ever traded 10 cents. What gives?
The bettor, however, needs to be compensated for the following:
1. The risk-free rate of return (e.g. parking the money in short-term bonds or other cash-equivalents, at around 4-5% year as of 2024).
2. Unforeseen risks resulting in the total loss of funds, that does not involve RFK somehow winning (e.g. regulatory risk, counterparty risk, exchange failure, etc.).
A year ago, it was not so improbable RFK could win. A lot can happen in a year. The ‘yes’ contract trading at 15 cents reflected both the non-zero probability of a win, plus the factors above. With RFK dropping out, the first probability goes to zero, but the other parts remain for the duration of the contract, so the contract does not become immediately worthless. Buying an RFK ‘no’ contract at 90 cents (or shorting the ‘yes’ contract at 10 cents, putting up $1 as collateral for each contract shorted) with a 12-month duration to expiration efficiently acts as a zero-coupon junk bond.
Therefore, buying the ‘no’ contract at 90 cents is not free money. Rather, it’s being compensated for the opportunity cost of having your money tied up until the contract expires vs. risk-free cash, and shouldering counterparty risk of losing everything. At the same time, shorting the ‘yes’ contract at 10 cents with a year until expiration is not free money either, as the opportunity cost is a risk-free bond yielding 5%, plus as mentioned above, also incurs counterparty risk. So you’re earning a 5% risk premium by having your money tied up for a year.
But if the ‘yes’ contract is trading at 3 cents with only 3 months until expiration, the effective yield is closer to 12%, which is a more enticing value proposition. So the ‘yes’ and ‘no’ contracts gradually exponentially converge to $0 and $1 respectively at a rate that compensates bettors for their risk. All of this is consistent in the framework of efficient markets. Or in other words, no free lunch.
In conclusion, prediction market prices will tend to imply artificially high probabilities for extremely improbable outcomes, but these should not be treated as probabilities, but rather as risk premium. Second, as we learned in 2016, polls should not be treated as probabilities, but at best crude approximations or sentiment. The question if polls are more accurate than markets is hotly debated. However, I think prediction markets are more accurate than polls, as they not only factor in public polling data, but other information as well, such as the Electoral College math, possible insider information by speculators, and turnout.